Estimation and inference in adaptive learning models with slowly decreasing gains